AI Glossary

Knowledge cutoff

Also called training cutoff

What does “Knowledge cutoff” mean?

Knowledge cutoff: The point in time after which a model's training data stops, so it has no built-in knowledge of later events. Coverage of the final months before the cutoff can be thin. Models connected to web search or retrieval can access newer information, but the model itself still doesn't know it on its own.

Knowledge cutoff visual guide: The point in time after which a model's training data stops, so it has no built-in knowledge of later events. Coverage of the final months before the cutoff can be thin. Models connected to web search or retrieval can access newer information, but the model itself still doesn't know it on its own.
Visual guide Open full size (new tab)

What is an example of “Knowledge cutoff”?

Without web access, a model may describe an older version of a product as the latest one because the newer release came after its cutoff.

Keep learning

See also

HallucinationTermRetrieval-augmented generationTermGroundingTermTraining dataTerm

Same topic

LatencyTermInference costTermMax tokensTermIn-context learningTermPrompt cachingTerm